The achievement by the open-source lab has sparked extensive discussion within the AI research community as it coincides with the prestigious annual Conference on Neural Information Processing Systems, or NeurIPS.
In its announcement on Monday, Hangzhou-based DeepSeek said V3.2-Speciale equalled Google’s Gemini 3 Pro, released two weeks ago, in reasoning capabilities, while V3.2-Speciale’s base model, the V3.2 introduced on the same day, performed on par with OpenAI’s GPT-5 launched in August.
DeepSeek said that V3.2-Speciale achieved gold-medal performance on the International Mathematical Olympiad test – an accomplishment previously reached only by internal models from OpenAI and Google DeepMind that have not been made public.
DeepSeek has open-sourced the V3.2 model on the developer platform Hugging Face. However, V3.2-Speciale is accessible solely through an application programming interface, or API, due to its “higher token usage”, according to the announcement.
